Changes in Plasma Metabolomic Profile Following Bariatric Surgery, Lifestyle Intervention or Diet Restriction—Insights from Human and Rat Studies

Although bariatric surgery is known to change the metabolome, it is unclear if this is specific for the intervention or a consequence of the induced bodyweight loss. As the weight loss after Roux-en-Y Gastric Bypass (RYGB) can hardly be mimicked with an evenly effective diet in humans, translational research efforts might be helpful. A group of 188 plasma metabolites of 46 patients from the randomized controlled Würzburg Adipositas Study (WAS) and from RYGB-treated rats (n = 6) as well as body-weight-matched controls (n = 7) were measured using liquid chromatography tandem mass spectrometry. WAS participants were randomized into intensive lifestyle modification (LS, n = 24) or RYGB (OP, n = 22). In patients in the WAS cohort, only bariatric surgery achieved a sustained weight loss (BMI −34.3% (OP) vs. −1.2% (LS), p ≤ 0.01). An explicit shift in the metabolomic profile was found in 57 metabolites in the human cohort and in 62 metabolites in the rodent model. Significantly higher levels of sphingolipids and lecithins were detected in both surgical groups but not in the conservatively treated human and animal groups. RYGB leads to a characteristic metabolomic profile, which differs distinctly from that following non-surgical intervention. Analysis of the human and rat data revealed that RYGB induces specific changes in the metabolome independent of weight loss.


Introduction
The prevalence of obesity has nearly tripled in the last 40 years and continues to increase in many parts of the world [1]. Recent studies report at least 30% of men and 35% of women to be obese in the US [2]. As a result, obesity and related metabolic disorders, particularly type 2 diabetes, are responsible for more deaths than undernourishment in most countries of the world [1]. Insulin resistance (IR), cardiovascular diseases and musculoskeletal disorders [3,4] as well as several types of cancer are well-known comorbidities of obesity [3,5,6]. Bariatric surgery, such as Roux-en-Y Gastric Bypass (RYGB), has proven not only to achieve a sustained weight loss but also a remission of diabetes and an improvement in other comorbidities [7,8].
A total of 188 metabolites were measured in 46 serum samples at randomization and one year after randomization. Forty-nine metabolites were not considered due to exclusion criteria. Thus, 139 metabolites (9 Cs, 21 amino acids, 9 biogenic amines, 13 lysoPCs, 72 PCs, a hexose and 14 SMs) were valid and retained for the main analysis and the derivation of metabolomic profiles. Fifty-seven metabolites were found to be significantly different between groups (1 acylcarnitine, 13 amino acids, 1 biogenic amine, 5 lysoPCs, 31 PCs, 5 SMs and a hexose) as shown in Table 2. Table 2. One-year follow-up means of 57 significantly different metabolites between LS and OP with standard deviation. p-values are adjusted for sex and age and corrected for multiple comparisons. C, acylcarnitine; lysoPC, lysophosphatidylcholine; PC aa, phosphatidylcholine with ester bonding; PC ae phosphatidylcholine with ether bonding; SM, sphingomyelin. The performed principal component analysis (PCA) resulted in twelve discriminating components with an eigenvalue > 1. The first two components explained 49% of the variance. Hence, further investigation was performed based on a two-component model. The first principal component (PC1) was characterized predominantly by phosphatidylcholines and sphingolipids, whereas the second principal component (PC2) was characterized by BCAA, aromatic acids and acylcarnitines (see Figure 1). The performed principal component analysis (PCA) resulted in twelve discriminating components with an eigenvalue > 1. The first two components explained 49% of the variance. Hence, further investigation was performed based on a two-component model. The first principal component (PC1) was characterized predominantly by phosphatidylcholines and sphingolipids, whereas the second principal component (PC2) was characterized by BCAA, aromatic acids and acylcarnitines (see Figure 1). We performed hierarchical clustering of the 57 metabolites with significant differences between the LS and OP group (see Figure 2) at one-year follow-up. A cluster of phosphatidylcholines, especially PC aa C42:Ys, as well as in SM (OH) C16:1, SM C26:1, lysoPC a C16:0, glutamine, glycine, citrulline and histidine were identified to be enriched only in patients from the OP group but not in patients from the LS group.  We performed hierarchical clustering of the 57 metabolites with significant differences between the LS and OP group (see Figure 2) at one-year follow-up. A cluster of phosphatidylcholines, especially PC aa C42:Ys, as well as in SM (OH) C16:1, SM C26:1, lysoPC a C16:0, glutamine, glycine, citrulline and histidine were identified to be enriched only in patients from the OP group but not in patients from the LS group.

Rodent Model
Characteristics of the rodent model have been described in detail previously [23]. Rats of the bariatric surgery group (RYGB_rat) experienced a significant weight loss of −7.3% (486.8 ± 9.3 g vs. 451.3 ± 18.5 g). In the body-weight-matched group (BWM_rat), a weight loss of 1.1% (481.9 ± 12.7 g vs. 476.4 ± 28.8 g) could be detected. At the time of the blood sampling, the body weights of RYGB_rat and BWM_rat were not significantly different (p = 0.16). A total of 188 metabolites were measured. Fifty-seven metabolites were excluded due to the below-mentioned criteria. A total of 131 metabolites (6 Cs, 19 amino acids, Figure 2. Heat map showing the scaled relative abundance of metabolites, which were significantly different between LS and OP group at one-year follow-up, visualized by hierarchical clustering. Abundance values represent row-wise z-scores of read counts. Each bar in the horizontal columns (which represent different patients) represents the expression intensity. Blue indicates a decreased level, red indicates an increased level. C, acylcarnitine; lysoPC, lysophosphatidylcholine; PC aa, phosphatidylcholine with ester bonding; PC ae phosphatidylcholine with ether bonding; SM, sphingolipid.

Rodent Model
Characteristics of the rodent model have been described in detail previously [23]. Rats of the bariatric surgery group (RYGB_rat) experienced a significant weight loss of -7.3% (486.8 ± 9.3 g vs. 451.3 ± 18.5 g). In the body-weight-matched group (BWM_rat), a weight loss of 1.1% (481.9 ± 12.7 g vs. 476.4 ± 28.8 g) could be detected. At the time of the blood sampling, the body weights of RYGB_rat and BWM_rat were not significantly different (p = 0.16). A total of 188 metabolites were measured. Fifty-seven metabolites were excluded due to the below-mentioned criteria. A total of 131 metabolites (6 Cs, 19 amino acids, 10 biogenic amines, 13 lysoPCs, 68 PCs, a hexose and 14 SMs) were valid and retained for the main analysis and the derivation of metabolomic profiles. Sixty-two metabolites were found to be significantly different as shown in Table 3. Heat map showing the scaled relative abundance of metabolites, which were significantly different between LS and OP group at one-year follow-up, visualized by hierarchical clustering. Abundance values represent row-wise z-scores of read counts. Each bar in the horizontal columns (which represent different patients) represents the expression intensity. Blue indicates a decreased level, red indicates an increased level. C, acylcarnitine; lysoPC, lysophosphatidylcholine; PC aa, phosphatidylcholine with ester bonding; PC ae phosphatidylcholine with ether bonding; SM, sphingolipid.  Further analysis of these metabolites (2 acylcarnitines, 6 amino acids, 2 biogenic amine, 3 lysoPCs, 41 PCs and 8 sphingolipids) with PCA showed that the first two components explained 74% of the variance. PC1 was mainly characterized by phosphatidylcholines. PC2 was not characterized by a specific subgroup of metabolites (see Figure 3) four weeks after the start of the interventions.
Further analysis of these metabolites (2 acylcarnitines, 6 amino acids, 2 biogenic amine, 3 lysoPCs, 41 PCs and 8 sphingolipids) with PCA showed that the first two components explained 74% of the variance. PC1 was mainly characterized by phosphatidylcholines. PC2 was not characterized by a specific subgroup of metabolites (see Figure 3) four weeks after the start of the interventions. We performed hierarchical clustering of the 62 metabolites with significant differences between the RYGB_rat and the BWM_rat group (see Figure 4). A cluster of phosphatidylcholines with ester and ether bonding, especially PC aa C42:Ys, as well as in SM (OH) C14:1, SM (OH) C24:1, lysoPC a C16:1, lysine, tryptophane, alanine and in the biogenic amines ADMA and SDMA were identified as being enriched only in the RYGB_rat group but not in the weight-identical BWM_rat group.  We performed hierarchical clustering of the 62 metabolites with significant differences between the RYGB_rat and the BWM_rat group (see Figure 4). A cluster of phosphatidylcholines with ester and ether bonding, especially PC aa C42:Ys, as well as in SM (OH) C14:1, SM (OH) C24:1, lysoPC a C16:1, lysine, tryptophane, alanine and in the biogenic amines ADMA and SDMA were identified as being enriched only in the RYGB_rat group but not in the weight-identical BWM_rat group.

Analysis of Overlapping Metabolomic Profiles in the Human OP and the Rat RYGB Group
Thirty-one metabolites with significant differences in both OP vs. LS and RYGB_rat vs. BWM_rat could be identified (see Table 4). Additionally, we performed an indirect comparison between the human data and the rodent model. In accordance with previously published data, the transferability of data from a morbidly obese patient to a rat with diet-induced obesity has to be performed with caution and only in cohesion with methodical comparability (e.g., comparable surgery) [24]. In indirect comparison to the conservatively treated groups LS and BWM_rat, the surgical groups OP and the RYGB_rat were characterized by metabolomic profiles of increased long-chain phosphatidylcholines and citrulline as well as decreased tryptophan (see Figures 5 and 6). The data are presented as means with standard deviation, respectively.
In terms of individual metabolites, six amino acids with a significant difference between RYGB_rat and BWM_rat were also found to be significantly different in the

Analysis of Overlapping Metabolomic Profiles in the Human OP and the Rat RYGB Group
Thirty-one metabolites with significant differences in both OP vs. LS and RYGB_rat vs. BWM_rat could be identified (see Table 4). Additionally, we performed an indirect comparison between the human data and the rodent model. In accordance with previously published data, the transferability of data from a morbidly obese patient to a rat with dietinduced obesity has to be performed with caution and only in cohesion with methodical  Table 4. Eighty-one significantly different metabolites subsequently in LS vs. OP of the WAS cohort at one-year follow-up and in animals treated with RYGB vs. diet-restricted animals of the rat cohort four weeks after start of the interventions. Thirty-one overlapping metabolites found to be significant in both comparisons are printed in bold. Arrows indicate if the metabolite is increased or decreased in the respective surgical group in comparison the conservative group. C, acylcarnitine; lysoPC, lysophosphatidylcholine; PC aa, phosphatidylcholine with ester bonding; PC ae phosphatidylcholine with ether bonding; SM, sphingolipid.

Discussion
The aim of our study was to identify and characterize a comprehensive metabolomic profile, which highlights the effect of bariatric surgery on the metabolome beyond weight loss. Whereas patients of the WAS intensified lifestyle group failed to lose weight significantly, patients with bariatric surgery lost a significant amount of their body weight as expected. To clarify the role of RYGB-induced body weight loss itself, the same metabolomic parameters were analyzed in rats treated with RYGB or a similar effective food restriction regime.
Examining the human data, twelve principal components with 57 significantly different metabolites between the LS and OP group were identified at 1-year follow-up. A characteristic metabolomic profile could be identified including two main components of differentiation. Upon closer examination of these components, sphingolipids and phosphatidylcholines were identified as the main metabolomic subgroups separating the two human cohorts. Both lipid groups are essential in the formation of the cell membrane and intracellular signal transduction.
Sphingolipids maintain lipid microenvironments of plasma membranes and form lipid rafts [19]. As well as their role as components of the cell membrane, sphingolipids build the carcass of myelin sheaths [25]. Recently, higher concentrations of sphingolipids together with low concentrations of BCAAs have been reported to correlate with a healthy liver phenotype [26]. The importance of sphingolipids in IR has been highlighted previously [27]. Phosphatidylcholines on the other hand build the major part of the membrane matrix [28]. Conclusively, a sensitive balance in these lipids is liable for polarization and signal transduction in the cell environment [29]. Recently, alterations in phosphatidylcholines and sphingolipids were identified as relevant, although differently regulated parts of the metabolomic profile in morbid obese as well as in patients with gastric adenocarcinoma and as key effectors of apoptosis and tumor cell growth, possibly explaining the well-known increased lifetime risk for cancer in obese subjects [30,31].
A further metabolomic subgroup included BCAA, and aromatic amino acids. The BCAAs could be identified as a valid separator of the LS and OP group at one year after randomization. As described before, BCAAs correlate with BMI and HOMA-IR [16,32,33].

Discussion
The aim of our study was to identify and characterize a comprehensive metabolomic profile, which highlights the effect of bariatric surgery on the metabolome beyond weight loss. Whereas patients of the WAS intensified lifestyle group failed to lose weight significantly, patients with bariatric surgery lost a significant amount of their body weight as expected. To clarify the role of RYGB-induced body weight loss itself, the same metabolomic parameters were analyzed in rats treated with RYGB or a similar effective food restriction regime.
Examining the human data, twelve principal components with 57 significantly different metabolites between the LS and OP group were identified at 1-year follow-up. A characteristic metabolomic profile could be identified including two main components of differentiation. Upon closer examination of these components, sphingolipids and phosphatidylcholines were identified as the main metabolomic subgroups separating the two human cohorts. Both lipid groups are essential in the formation of the cell membrane and intracellular signal transduction.
Sphingolipids maintain lipid microenvironments of plasma membranes and form lipid rafts [19]. As well as their role as components of the cell membrane, sphingolipids build the carcass of myelin sheaths [25]. Recently, higher concentrations of sphingolipids together with low concentrations of BCAAs have been reported to correlate with a healthy liver phenotype [26]. The importance of sphingolipids in IR has been highlighted previously [27]. Phosphatidylcholines on the other hand build the major part of the membrane matrix [28]. Conclusively, a sensitive balance in these lipids is liable for polarization and signal transduction in the cell environment [29]. Recently, alterations in phosphatidylcholines and sphingolipids were identified as relevant, although differently regulated parts of the metabolomic profile in morbid obese as well as in patients with gastric adenocarcinoma and as key effectors of apoptosis and tumor cell growth, possibly explaining the well-known increased lifetime risk for cancer in obese subjects [30,31].
A further metabolomic subgroup included BCAA, and aromatic amino acids. The BCAAs could be identified as a valid separator of the LS and OP group at one year after randomization. As described before, BCAAs correlate with BMI and HOMA-IR [16,32,33]. Studies analyzing human adipose and muscle tissue in obese and diabetic conditions show a reduced activity of mitochondrial branched-chain aminotransferase. This is in accordance with the development of IR, especially if the concentration of BCAAs is elevated over a longer period of time [34]. Subjects with chronically elevated concentrations of BCAAs also have higher HOMA-IR values. Elevated leucine and valine concentrations have been associated with IR [35]. Among others, BCAAs affect the intracellular insulin signaling in human cells by binding to Rag GTPases, leading to the activation of mammalian target of rapamycin complex 1 (mTORC1), the activation of ribosomal S6 kinase 1 (S6K), the inhibition of insulin receptor substrate 1 (IRS-1) and the stimulation of mitochondrial dysfunction [36]. See Supplementary Figure S2 for further explanation.
Analyzing overlapping phosphatidylcholines from the metabolomic profiles as a translational approach with indirect comparison, PC aa C42:Ys have been identified as being significantly different metabolites in both comparisons, OP vs. LS and RYGB_rat vs. BWM_rat. PC aa C42:Ys, better known as lecithins, have been repeatedly mentioned in studies investigating metabolomics in obesity [16,24,28,37]. With respect to the method of analysis, the length of the two fatty acids on these phosphatidylcholines could not be delineated. However, the most frequently identified PCs are arachidic acid, lignoceric acid and arachidonic acid. All three of these fatty acids have been reported to be decreased in obese patients [12,38,39]. In the present cohort, a significant enrichment of PC aa C42:Ys was found in both surgical groups in the human and rat cohort. Interestingly, no significant increase in lecithins was detected in both conservatively treated groups. Therefore, an effect of RYGB on lecithins independently of weight loss can be assumed. As the weightloss-independent effect of RYGB is a recent focus of clinical research to find effective and sustainable alternatives for the therapy of morbidly obese patients, the presented changes in the metabolome after bariatric surgery, consensually, is in accordance with previous findings [10,40].
The analysis of the metabolomics of rats with similar body weights after RYGB and diet restriction and similar food composition overcomes the limitation that weight loss effects cannot be differentiated from the effects of the bariatric intervention (i.e., changed absorbance) itself. Interestingly, lipids play a crucial role in differentiating RYGB from BWM rats, while amino acids do not facilitate the separation of these intervention groups. Therefore, changes in BCAAs after RYGB do not seem to be caused by bariatric surgery itself. A significant increase in sphingolipids as well as PC aa C42:Ys in the OP group of the WAS trial and higher levels in the BWM vs. RYGB group of our animal experiment were detected. The role of sphingolipids in obesity and IR is not fully understood and has to be investigated in further human studies with an adequate diet restriction paradigm [41].
Our study has several limitations. The very low sample size compared to the large number of analyzed metabolites may have hampered the interpretation of data. It was also not possible to consider the whole spectrum of confounders (e.g., smoking status) potentially influencing the metabolic spectrum. Although there was no difference in the nutrition counseling, we cannot exclude that the actual nutrition might have differed from the references within the nutrition counseling. Additionally, gender distribution differed significantly between the LS and OP group of the WAS cohort. Furthermore, a direct comparison of human metabolomic data with the metabolome of rats is not uncritical and therefore allows only limited conclusions regarding the underlying pathophysiological mechanism [42,43]. Nevertheless, the strength of our study was the inclusion of excellent annotated material, both from humans and animals that have undergone a randomly assigned intervention.

Patients
The Würzburg Adipositas Study is a randomized trial comparing the effects of RYGB vs. psychotherapy-enhanced lifestyle intervention not including calorie-limited nutrition in morbidly obese patients. Details of the design of the study are published elsewhere [22]. Twenty-four participants were randomized into intensive lifestyle modification, 22 were randomized into an RYGB-surgery group, which was performed at the department of general, visceral, transplant, vascular and pediatric surgery in the University Hospital of Wuerzburg. At twelve months after randomization, blood samples were taken for further analyses (see Figure 7). The WAS study protocol was approved by the Ethics Committee of the University Hospital of Wuerzburg (182/08). All patients provided written informed consent.

Patients
The Würzburg Adipositas Study is a randomized trial comparing the effects of RYGB vs. psychotherapy-enhanced lifestyle intervention not including calorie-limited nutrition in morbidly obese patients. Details of the design of the study are published elsewhere [22]. Twenty-four participants were randomized into intensive lifestyle modification, 22 were randomized into an RYGB-surgery group, which was performed at the department of general, visceral, transplant, vascular and pediatric surgery in the University Hospital of Wuerzburg. At twelve months after randomization, blood samples were taken for further analyses (see Figure 7). The WAS study protocol was approved by the Ethics Committee of the University Hospital of Wuerzburg (182/08). All patients provided written informed consent.

Animals
As described in detail elsewhere, adult male Wistar rats (Charles River Laboratories, n = 13) with initial body weight 323.1 ± 4.7 g, 9-10 weeks old, were group-housed in a certified facility with an ambient room temperature of 22 °C and a 12-h light/dark cycle. These animals were part of a series that was recently published [44]. Animals had free access to a high-fat diet (C1090-60 HF diet, 5228 kcal/kg; 60% calories from fat, 16% from protein and 24% from carbohydrate; Altromin, Lage, Germany) for about 6 weeks to induce obesity. The animals were then randomized into the following treatment groups and, after the respective intervention, kept on a choice of the high-and a low-fat diet (C1090-10 LF, 3514 kcal/kg; 10% calories from fat, 24% from protein and 66% from carbohydrate; Altromin). RYGB was performed on six animals. Seven animals were body-weightmatched controls, which underwent sham surgery and were then kept on chronic food restriction to induce a similar weight course as in RYGB animals. This was achieved by restricting the amount of high-and low-fat diet they consumed compared to that of RYGB_rat. As published before, animals were isoflurane-anesthetized and under butor-

Animals
As described in detail elsewhere, adult male Wistar rats (Charles River Laboratories, n = 13) with initial body weight 323.1 ± 4.7 g, 9-10 weeks old, were group-housed in a certified facility with an ambient room temperature of 22 • C and a 12-h light/dark cycle. These animals were part of a series that was recently published [44]. Animals had free access to a high-fat diet (C1090-60 HF diet, 5228 kcal/kg; 60% calories from fat, 16% from protein and 24% from carbohydrate; Altromin, Lage, Germany) for about 6 weeks to induce obesity. The animals were then randomized into the following treatment groups and, after the respective intervention, kept on a choice of the high-and a low-fat diet (C1090-10 LF, 3514 kcal/kg; 10% calories from fat, 24% from protein and 66% from carbohydrate; Altromin). RYGB was performed on six animals. Seven animals were body-weight-matched controls, which underwent sham surgery and were then kept on chronic food restriction to induce a similar weight course as in RYGB animals. This was achieved by restricting the amount of high-and low-fat diet they consumed compared to that of RYGB_rat. As published before, animals were isoflurane-anesthetized and under butorphanol (0.1 mg/kg) analgesia for RYGB and sham operation [45]. For RYGB, a small gastric pouch 5% of the original stomach size was created, and the biliopancreatic and common limbs were made to measure 15 cm and 25 cm in length, respectively [45,46]. Four weeks after intervention, blood samples were taken in a fasted state (12 h). The local regulatory authority (Regierung von Unterfranken: 55.2-2532-2-467) approved all animal procedures.

Laboratory Measurements
In brief, serum samples from patients were obtained upon enrolment under standardized conditions. After an overnight fast, the blood was drawn from the cubital vein with a Safety-Multifly ® 21G (Sarstedt AG & Co.KG, Nümbrecht). Samples for analysis of fasting glucose were stored in S-Monovette ® Fluoride/EDTA (Sarstedt AG & Co.KG). Samples for the analysis of insulin and metabolome were stored in S-Monovette ® Serum (Sarstedt AG & Co.KG). All samples were centrifuged immediately after blood drawn at 500× g for 5 min. Afterwards, serum was stored at −80 • C until further analysis. HOMA-IR was calculated as follows: HOMA-IR = insulin (fasting) (mU/l) * glucose (fasting) (mg/dl)/405 Plasma samples from rats were collected after a 12-h fasting period under deep anesthesia shortly prior to euthanasia. Immediately after collection in tubes pretreated with a DPP-IV inhibitor (Merck), plasma was separated from the blood samples by centrifugation at 5 krpm for 10 min at 4 • C and stored at −80 • C.
for PCs, hexose, Cs and SMs. In this dataset, metabolites with more than 60% measurements at the limit of detection (LOD) or below lowest limit of quantification (LLOQ), and those samples with missing values for more than 60% of the metabolites were excluded from the analyses as described previously [51]. To ensure comparability of data between batch measurements, each metabolite value was normalized to four human reference samples included into each batch as previously described. Normal ranges for metabolites based on data described elsewhere [25,26]. In the nomenclature of presented lipids Cx:Y' describes the composition of the lipid chain with x' indicating the number of carbon atoms and y' the number of double bonds.

Statistical Analysis
Analyses were conducted using SPSS software (PASW version 25.0, SPSS Inc. Chicago, IL, USA), GraphPad Prism version 9.1.2 for Windows (GraphPad Software, La Jolla, CA, USA) and MetaboAnalyst (version 5.0, www.metaboanalyst.ca, accessed on 10 October 2022). The dataset was tested for normal distribution using the Shapiro-Wilk test. All normally distributed values are presented as mean ± SD whereas non-normally distributed data are presented as medians ± interquartile range (IQR). Outliers were defined with values >1.5 interquartile ranges (IQR) below the first quartile or above the third quartile. No outliers were identified in the human cohort. One rat was identified as an outlier and excluded from further analysis. Post hoc comparisons (Tukey-Kramer Honest Significant Difference method) were performed to control for family-wise error rates (FWER). Hierarchical clustering analysis was conducted on the final dataset to assess their putative abundances (clustering by Euclidean distance measure and Ward linkage). To visualize class differences from a multivariate dataset and to determine whether there is any cluster distinction between different groups, principal component analysis was performed. The validity of the PCA was calculated using Bartlett's test and the Kaiser-Meyer-Olkin measure of sampling adequacy. According to the Kaiser criteria, only factors with an eigenvalue ≥ 1 were considered. The calculated principal components were defined as factor analysis components.
Clinical data of the WAS cohort was not normally distributed (Shapiro-Wilk test, p < 0.05). Hence, we performed Mann-Whitney tests for clinical parameters to investigate differences between groups at one-year follow-up. Data from the metabolomics followed normal distribution (Shapiro-Wilk test, p > 0.05). To adjust for differences at baseline, a repeated-measures analysis of covariance (ANCOVA) was performed (correcting for age and sex).
Data in the rodent model were normally distributed (Shapiro-Wilk test, p > 0.05). A one-way ANOVA was performed to assess the effects of obesity treatment on body weight and the metabolome four weeks after intervention. p-values for the respective comparisons after correction for multiple comparisons are reported.

Conclusions
We analyzed the effects of surgical vs. lifestyle intervention on the serum metabolome of obese and insulin resistant human patients of the randomized controlled WAS and compared them with the results from a rodent model. As the conservatively treated human group failed to achieve significant weight loss, the difference in the metabolomic profile could either be associated with RYGB or weight loss. However, several metabolites were found to be significantly different in rats following RYGB vs. rats treated with an evenly effective food restriction paradigm. By indirectly comparing the metabolomic profiles of both species as a translational approach, a high number of overlapping parameters could be found. A subgroup of sphingolipids, BCAAs and phosphatidylcholines was found to be altered in both surgical groups, but not in the conservatively treated human and rat groups. Thus, these metabolites might indeed be a specific consequence of the RYGB. The significant increase in the hereby-identified lipids as well as a significant decrease in BCAA has been shown before to improve the glucose metabolism. Thereby, our investigation is in accordance with several methodically comparable studies, clarifying the complex role of RYGB, and, hence, supports the worldwide effort to investigate the weight-lossindependent effects of bariatric surgery.
Supplementary Materials: The following supporting information can be downloaded at: https://www. mdpi.com/article/10.3390/ijms24032354/s1, Figure S1: Unsupervised principal component analysis of 13 significantly different metabolites between lifestyle intervention and surgery group in the WAS cohort at baseline visit.; Figure S2: Intracellular BCAA pathway leading to insulin resistance.